In order to solve the problem that the calculation accuracy and the stability of the classical response surface method is not high in some cases, an improved response surface method based on the cross order reduction method for structural reliability analysis was proposed. First of all, the way of the selection of initial point was improved, Secongly, in the process of selecting sampling points, the star design method was introduced in this paper, and the n-dimensional coordinate system was divided into n-1 one-dimensional coordinate systems without sampling center points and one one-dimensional coordinate system with both sampling center points and experiment points. In the former n-1 coordinate systems, the distance between the sampling points and the actual limit state surface was considered to select the sampling point, while in the last coordinate system, a better point was selected between the positive experiment point and the sampling center point, and then the selected n sample points were used to fit the linear response surface function until the convergence condition was met. Finally, the feasibility of the proposed method is verified by comparing with traditional response surface method and weighted regression method, and the rusults are presented to demonstrate that the proposed method guarantees higher accuracy and can ensure its computational efficiency and stability. |